Money Machine: Do the Poor Demand Clientelism?
Transcript of Money Machine: Do the Poor Demand Clientelism?
Working Paper No. 14 2017
The Program on Governanceand Local Development
at Gothenburg
The Program on Governanceand Local Development
Kristen Kao, Ellen Lust and Lise Rakner
Money Machine: Do the Poor Demand Clientelism?
Cover Photo: Political rally in Malawi taken by Kim Dionne
Kristen Kao (University of Gothenburg, GLD)
Ellen Lust (University of Gothenburg, GLD)
Lise Rakner (University of Bergen)
1
Abstract
The literature on clientelism suggests that the poor are particularly likely to exchange their votes
for cash or material goods. In this supply-side perspective, candidates are more likely to offer
goods in return for votes to the poor because the poor sell their votes at a lower price, are more
likely to act reciprocally, and are less likely to see vote-buying as morally unacceptable. We know
much less about the poor’s demand for vote-buying. Studies suggest that the middle class punishes
vote-buying candidates, but assume that the poor welcome offers. Employing a rating-based,
conjoint analysis in Malawi to examine the poor’s preferences over vote-buying, we find that the
poor are repelled by candidates who promise an immediate exchange of particularistic goods for
votes and prefer candidates who promise community goods. This highlights the need to consider
the possibility that candidates incur costs when offering to buy votes in poor communities.
2
1. Introduction
The literature on democracies in the developing world paints a picture of rampant vote-buying in
which the poor, in particular, vote for candidates in exchange for direct, tangible benefits in
elections, rather than campaign promises of public goods or national legislation (Dixit and
Londregan 1996; Stokes 2005; Blaydes 2011; Jensen and Justesen 2014; Kitschelt 2002; Weitz
Shapiro 2012). In large part, this view is based on an argument that candidates target such vote-
buying efforts to the poor because the poor sell their vote at a lower price, are more likely to act
reciprocally, and are less likely to see vote-buying as morally unacceptable. Yet, it does not follow
that the poor prefer such offers. The lack of a correlation between campaign expenditure and
electoral outcomes (Bjørkman, 2013) suggests, at least indirectly, that vote-buying may not impact
balloting to the extent believed. Moreover, recent studies of African voting behavior suggest that
political performance and issue-based campaigns may be more important to voters than ethnicity
(Ferree 2010, Lindberg and Morrison 2008, Bleck and van de Walle 2011, Carlson 2015). Money
may flow freely at election time,1 but is the exchange of goods for votes what citizens prefer?
In this paper, we examine voters’ preferences and find evidence that the poor may actually be
repulsed by candidates who seek their votes through short-term, self-interested incentives. We
employ a rating-based, conjoint analysis (Hainmueller et al. 2013) embedded in the 2016 Local
Governance Performance Index (LGPI)2 survey of over 8,100 Malawians (Lust et al. 2016). The
research design is novel and - we believe - the first attempt at applying conjoint analysis to
understand vote-buying. Employing a conjoint survey experiment in which respondents are asked
to rate the likelihood of voting for a candidate with randomly varied clientelistic appeals, provides
1 Mares and Young (2016) report that 16% of voters surveyed in 33 countries during the fifth round of the Afrobarometer, and 15% of those in Latin America in 2010 and 2012 rounds of the Americas Barometer, reported being offered money or goods in exchange for their vote during the last election. List experiments aimed at overcoming social desirability bias find even higher rates of vote-buying. An estimated 1 in 4 (24%) of Nigerians accepted compensation in return for their vote. Gonzalez-Ocantos et al, 2012. 2 To find out more information about this survey, please visit www.gld.gu.se.
3
an opportunity to examine the poor’s ‘pure’ preferences over vote-buying. In the real world, of
course, candidates present a bundle of appeals -- declaring that they will deliver roads, water, and
health clinics as they pass out bags of sugar and rice. The conjoint experiment allows us to weigh
the relative importance of such appeals in determining voters’ choices. Employing this in Malawi
is particularly useful for examining the assumptions that the poor welcome vote-buying. Malawi
has one of the poorest populations in the world.3 If poverty prompts individuals to accept voter-
sellers’ offers, Malawian voters should respond favorably to these incentives. The Malawian case
can spur our thinking about vote-buying and poverty elsewhere, advancing the literature on
clientelism in developing democracies by explicitly considering the demand-side of clientelism.
We find evidence that voters are driven by community interests, not short-term, targeted
incentives. Malawians respond most favorably to a promise of community goods, followed by a
promise of future, personal assistance. They respond less favorably to those who promise
immediate exchanges of tangible goods for votes, as emphasized in much of the current literature.
In short, Malawians find vote-buying much less appealing than scholars often assume; rather, they
support candidates who promise public goods for their area.
This paper proceeds as follows. It begins by examining the literature on clientelism, pointing to
how scholars have failed to explore whether the poor prefer campaigns centered on handouts of
money, sugar, rice and other material goods. Second, we outline the political context in Malawi
focusing on why Malawi is a particularly useful case to examine and on how the political context
affected design choices in the experiment. The third section presents the findings in the conjoint
experiment and its results. A fourth section discusses our findings, drawing on focus group
discussions conducted to substantiate our experimental results and considering alternative
explanations. A final section considers the implications for future studies of clientelism.
3 According to World Bank data, Malawi is the fourth poorest country in the world based on GDP per capita (WB 2017)
4
2. Do the Poor Prefer Vote-Buying?
Despite extensive literature on clientelism,4 there has been remarkably little attention paid to the
choices that voters make -- when and where they accept the clientelist bargain, and when they turn
away from it (for an exception, see Pellicer et al, 2016). Scholars who examine an immediate
exchange of a vote for a tangible good, or vote buying, emphasize candidates’ strategies. They
consider when and how candidates decide to purchase votes (Weitz Shapiro 2012), whether they
do so from swing voters or hard-core supporters (Stokes et al. 2013, Cox and McCubbins 1986,
Grossman and Helpman 1996, Corstange 2017, Justesen and Manzetti 2017), and if candidates
aim to change individuals’ votes or to mobilize them – both at the polls (Nichter 2008, Kramon
2015: 24) and in campaigns (Munzu 2014). They also look at how parties engage vote brokers in
their efforts, and these vote brokers’ strategies (Stokes et al. 2013, Finan and Schechter 2012,
Szwarcberg 2012, 2015). The question is not whether voters are willing to sell their vote, but rather
which voters are given the opportunity to do so.
The studies overwhelmingly anticipate that candidates target the poor (e.g. Diaz-Cayeros et al.
2016, Corstange 2016, Stokes et al. 2013, Weitz Shapiro 2012). Candidates may be more likely to
target the poor and less educated because their votes are ‘cheaper,’ and thus may benefit from the
poor as an inexpensive voting bank (Dixit and Londregan 1996, Magaloni 2006, Blaydes 2011,
Stokes 2005, Jensen and Justesen 2014). Second, the poor may be more likely to display norms of
trust, reciprocity and caring, something that both vote brokers and voters can capitalize upon in
order to solve their commitment problem (Auyero 1999, 2000, Finan and Schechter 2012). These
studies do not necessarily assume that the poor have intrinsic preferences for vote-buying. We can
understand these as induced preferences if the poor are in environments where these are the offers
candidates make. Importantly, however, this literature overlooks the possibility that if the poor
dislike vote-buying, then candidates may incur costs by presenting such offers.
4 We define “clientelism” as the quid-pro-quo exchange between a patron and a client in which the patron trades a targeted good, service, or favor for political support from the client (see Hicken (2011) and Mares and Young (2016) for reviews).
5
A second strand of the literature reasons that the poor actually prefer vote-buying. First, drawing
from the large literature in behavioral economics which finds that the poor exhibit present-bias
(Carvalho et al. 2016, Haushofer and Fehr 2014, Lawrence 1991, Shah et al. 2012, Tanaka et al.
2016), scholars of clientelism argue that the poor are likely to prefer tangible handouts in the
present over redistributive benefits in the future (Deposato 2007; Kitschelt 2000; Scott 1969).
According to Scott (1969, 1150), “Poverty shortens a man’s time horizon and maximizes the
effectiveness of short-run material inducements.” Or, as Kitschelt (2000, 857) put it more recently:
“poor and uneducated citizens discount the future, rely on short causal chains, and prize instant
advantages such that the appeal of direct, clientelist exchanges always trumps that of indirect,
programmatic linkages promising uncertain and distant rewards to voters.” Second, the poor may
also be more likely to see vote-selling as acceptable, either because vote-buying is so frequently
exercised in their communities that this behavior comes to be seen as appropriate, or because they
lack the ability to “observe, understand, and believe the collective downsides of vote buying”
(Gonzalez-Ocantos et al. 2013, 198). Finally, it may not be the material goods themselves, but the
signal they send, that explains why the poor prefer candidates who offer to buy their votes. In a
study of vote-buying in Kenya, Kramon (2016, 33) concludes that, “Among the poorest
participants, information about vote buying generated positive expectations of the candidate’s
future performance.” These scholars suggest that a preference for vote-buying exists. The poor
are expected to embrace vote-buying, in contrast to the middle class, which Weitz-Shapiro (2014)
argues punishes candidates employing clientelism.
Yet, studies from social psychology suggest that the poor may not prefer vote-buying; indeed, they
may be even repulsed by it. They find the poor to be more likely than the wealthy to prioritize
community needs over individual incentives. The poor tend to focus on the choices, conditions
and needs of others (Dietze and Knowles 2016, Piff and Robinson 2017, Stephens et al. 2007,
Stellar et al. 2012) and, perhaps due to their greater need for assistance, to develop a greater
propensity toward altruism than the wealthy (e.g., Piff et al. 2010). Moreover, they more often view
success (and failure) as the result of structural conditions rather than individual skill and hard work
6
(Krause et al. 2009, Piff et al. 2010), and consequently, may place greater value on improving health
care, education and other conditions in the community. If vote-buying and community goods are
seen as trade-offs, the poor may not prefer candidates offering particularistic incentives. Similarly,
behavioral economists find that monetizing exchange often undermines reciprocal relationships
(Bowles 2016). There are, of course, reasons to question whether these experimental studies,
generally conducted on students from Western, educated, industrialized, rich and democratic
(WEIRD) societies, inform our understanding of the actual poor’s preferences for vote-buying in
less wealthy, developing democracies of the global South (Henrich et al. 2010). However, the same
studies also give reason to question the untested assumption that the poor prefer vote-buying or, at
least, are not repelled by it.
To interrogate the poor’s preferences toward vote-buying, we exploit a survey experiment
conducted in Malawi in 2016. The experiment, described in more detail below, was designed to
examine the extent to which respondents preferred parliamentary candidates who offered
immediate targeted goods, targeted goods in the future, or community benefits. The experiment
helps decipher preferences over vote-buying in the absence of direct questions, thus avoiding
social desirability bias associated with vote-buying.5 It also helps us to examine the poor’s pure
preference for vote-buying, isolated from other, simultaneously presented appeals. Segmenting
respondents by level of wealth according to a number of different indicators, we leverage the
experiment to explore the expectation that the poor are more likely to prefer those offering them targeted
incentives over those offering public/club goods for their communities.
3. Malawi’s Political Context
Malawi provides a useful case for analyzing the effects of poverty and vote-buying. It is one of the
poorest countries in the world, ranking 173 out of 188 countries in the Human Development
5 Surveys in five Latin American countries find that vast majorities of respondents disapproved or disapproved strongly of vote buying (Gonzalez-Ocantos et al. 2014).
7
Index. Since 1993, the country has held uninterrupted multiparty elections every five years with
rather high participation; the last national elections held in 2014 had a turnout rate of 71 percent.
Yet, the legislative system is weak and highly personalized (Dulani and Donge 2005). Arguably,
Malawi constitutes an ideal setting in which voters should be expected to prefer candidates who
offer them targeted incentives. It thus is a ‘hard case’ for challenging the expectation that the poor
prefer candidates who offer to buy their votes.
Malawi’s elections are generally free, fair and competitive, but voters electing members of
parliament (MP) recognize that the parliament is constrained. Executive powers are strong in
Malawi, and parliamentary oversight is generally weak (Dulani and Donge 2005; Wahman and
Patel 2015). The parliament does not control its own budget or determine when it convenes and
for how long it should meet. This political arrangement discourages parliamentarians from
pursuing agendas of broad-sweeping national legislation in favor of more targeted pork-barrel
benefits for their districts or particularistic benefits for individual constituents6 (Dionne and Dulani
2013, Rakner and Svåsand 2013, Wahman and Patel 2015). Consequently, we do not expect voters
to be influenced by candidates’ broad policy positions at the national level.
However, MPs do have significant power over the provision of local public goods (Ejdemyr,
Kramon, and Robinson 2015). In 2006, legislation gave MPs access to Constituency Development
Funds for development projects in their district as well as discretion over which projects would
receive funds. The amount of these funds has increased from about 2,500 USD per constituency
in 2006/2007 to 17,500 USD per constituency in 2013 (The Nation 2013), in a country where the
per capita GDP is about 227 USD per capita (World Bank 2015). This greatly increases MPs’ ability
to provide goods to their constituents’ communities. Candidates’ promises to improve health,
education, and other local services are credible. We thus include such promises as one treatment
arm in the experiment.
6 To illustrate, in the words of a Malawian female MP: “Without handouts your people in your constituency can take you as a useless parliamentarian. Whenever I am at home, I spend almost Kw 100.000 (USD 239) in handouts every day (Nyasa Times 2014).
8
In the lead-up to elections in Malawi, vote-buying is common. There is no regulation of campaign
finance, and the practice of vote buying occurs on a large scale, involving politicians from all parties
as well as traditional leaders (Ballington and Kahane 2014, Mpesi and Muriaas 2012). Beginning
with the first democratically elected President Muluzi, Malawi has witnessed candidates openly
providing monetary and material items at political rallies.7 According to Tambulasi and Kayuni
(2002, 151), the culture of hand-outs of cash and other material goods at times of election is an
attempt to erase the forced gift legacy of the former autocratic president-for-life Banda, where
citizens were expected to provide gifts to the President. Regardless of the reason, candidates’
widespread and open distribution of cash is widely acknowledged. This was reflected in focus
groups discussions that we conducted before we carried out the LGPI, which informed the items
used as selective incentives in the experiment.8
There is also widespread evidence of targeted incentives being provided after the elections as part of
a clientelistic bargain. Respondents in the focus groups, as well as the scholarly works, report that
subsidized fertilizer, and Food for Work Programs are distributed to political supporters (Banful
2011, Patel, and Wahman 2015)9. Other services mentioned include buying coffins for funeral
services, providing transport to members of the constituents, introducing football trophies,
providing milling services, and distributing of financial aid. Respondents who are asked to rate
their likelihood of preferring candidates who offer future selective incentives are thus likely to see
these candidates as realistic.
7 The open distribution of cash witnessed in Malawi has been commented on by election observer teams from the African Union expressing their surprise with the way people at political rallies were given handouts openly (The Nation 2004). See also the report from European Commission’s Observer mission in 2014 (EU 2014). 8 See Online Appendix E. In Focus group discussions, respondents reported that candidates offered them gifts such as salt, money, sugar, and clothes as part of their campaign strategies. 9 Focus group discussions prior to the implementation of our survey revealed that politicians were considered to target services and benefits to members of their constituencies (see Online Appendix E).
9
Moreover, Malawians not only expect local public goods as well as particularistic incentives, but
they prefer that their MPs concentrate on providing these types of goods. In the 2016 LGPI, we
asked a sample of 1,461 randomly selected respondents which of the following activities they
would like current MPs to prioritize from among soliciting funds to finance projects in your
constituency (for example, funds for hospitals or schools), passing national laws, or providing
individual services to citizens of your constituency (helping them to obtain licenses, find jobs,
educate children, etc.). We find that 56 percent of the respondents prefer for their MPs to focus
on financing projects, while 27 percent want them to provide individual services, and only 12
percent wanted them to spend their time passing national laws.10 Thus, concentrating on voters’
preferences over immediate selective incentives (e.g., cash, sugar), future selective incentives (e.g.,
fertilizer subsidies, assistance with services), and community goods (e.g., schools, clinics) reflects
the reality of Malawi’s political system and voters’ expectations.
4. The LGPI and Experimental Design
We exploit a conjoint analysis in order to examine the poor’s attitude toward candidates who
engage in vote-buying. Conjoint experiments are widely employed in market research to
understand the preferences of consumers, but they have recently become popular in the field of
political science for understanding the preferences of voters over candidates (Franchino and
Zucchini 2015).
Conjoint experiments have a number of advantages. By showing respondents a set of candidate
characteristics rather than asking them to directly evaluate each characteristic by itself, the
experimental design mitigates issues of sensitivity and helps researchers to overcome social
desirability bias (Hainmueller and Hopkins 2015). Additionally, conjoint analysis overcomes
limitations that exist with more traditional framing or vignette experimental methods used to
10 These figures include 3 percent of respondents who wanted something else and 2 percent who said they did not know.
10
understand voter behavior. Typically, such experiments involve varying just one candidate
characteristic at a time to ensure that the researcher is able to isolate the effect of that characteristic
on participant behavior. This design requires a very large number of respondents for multiple
characteristics, does not allow the researcher to examine the effect of interactions between
candidate characteristics, and fails to help the researcher understand multidimensional decision
making since only one dimension is varied at a time. Conjoint experiments overcome these
problems, allowing for more characteristics to be altered, and to be varied independently from
other attributes. In reality, candidates are likely to promise the provision of multiple types of goods
at once, making it difficult to know the differential effects of each one in shaping voter preferences.
Although it sacrifices external validity, this experimental design allows us to artificially isolate the
effects of one type of good compared to another.
We employed a rating-based single vignette experimental design in this study. In single vignette
experiments, respondents are presented with, and asked to evaluate, a single candidate. This design
is viewed as less powerful than paired comparisons, in which respondents compare two candidates
side-by-side (Hainmueller, Hangartner, and Yamamoto 2015). In this case, however, it provides a
superior design. Individuals are not comparing candidates who engage in vote buying vs. those
who do not, but rather are simply asked to rate the appeal of a single candidate who provides
community goods or targeted incentives. By doing so, we are less likely to cue respondents to
focus on vote-buying, thus avoiding social desirability bias.
The experiment was embedded within the LGPI. The LPGI is a national face-to-face survey aimed
at understanding individuals’ experiences, satisfaction and perceptions of governance and service
provision. We implemented the survey in March of 2016, using tablet computers. The experiment
was seen by a random subsample of 1,191 of the survey respondents.11
The stem of the question that all respondents received read as follows: “I am about to read you
11 See http://gld.gu.se/en/research-projects/malawi-project/ for more information on this survey. The total number of participants in the survey was approximately 8,100.
11
the descriptions of a candidate for parliament. Then I will ask you how likely you would be to vote
for this parliamentary candidate.”12 The interviewer then read a description of a candidate and
asked the respondent: “How likely is it that you would vote for this parliamentary candidate: very
likely, somewhat likely, not likely, not at all likely.” To assure that the respondents had the
opportunity to consider the information fully, the interviewer also offered to read the description
of the candidate again. Once the respondent indicated that he or she was ready to answer, the
interviewer recorded the answer.
The experimental setup involves randomly altering the candidate characteristics. The random
assignment of profile characteristics in conjoint analysis allows for the testing of numerous
candidate characteristics at once, while maintaining a low number of respondents. For this study,
candidate characteristics varied in a number of ways including campaign appeals, co-ethnicity with
the respondent, and strong (weak) ties to the community. These characteristics are summarized in
Table 1.13
12 We mention parliament twice to ensure that the respondent is thinking about the correct election and not the local or presidential elections that take place at the same time as the parliamentary elections. 13 Wording of the statement read to respondents is provided in Online Appendix A.
12
Table 1. Randomized Attributes and Characteristics for Hypothetical Candidates
Characteristic Choices
Campaign Appeal14 � Immediate Targeted Goods – bags of sugar, salt, and
K500 bills.
� Future Targeted Goods – promise of fertilizer
subsidies/financial aid for funerals, and help with
personal problems.
� Future Public Goods – promise of more schools,
improved healthcare, digging of boreholes.
Co-Ethnicity15 � Co-ethnic
� Non-co-ethnic (multiple)
Strong/weak ties to the
community
� Born in the village/ward, has lived in the area for a
long time
� Has recently moved to back to this village/ward after
many successful years living abroad
14 We also included negative campaign appeals in this experiment, by which we mean criticism of candidates who offer immediate selective goods, criticism of candidates who offer future targeted goods, and criticism of candidates who offer future club goods. We analyze these further in other work. We present the full results in Online Appendix B. The inclusion of negative campaign appeals does not alter results. 15 Although ethnic and regional identities are politically salient in Malawi (Kaspin 1995, Posner 2004), in keeping with the focus of this paper on vote-buying and clientelism, the analyses we present below do not report the effects of co-ethnic and co-local identities. A paper examining the impact of co-ethnic versus co-local identities on the likelihood of vote for a candidate is in the process of being written. However, full results are included in Online Appendix B for those interested in these aspects of the experiment. The inclusion of identity traits does not alter the results reported here.
13
Targeted Incentives and Community Goods. The primary goal of the analysis presented here is to test
whether poor voters prefer candidates who make clientelistic appeals, including short- and long-
term selective incentives, or community-oriented public goods. To do so, we consider how
respondents evaluated candidates making different campaign appeals. The experiment presented
six possible candidate platforms, three of which are the focus of this analysis: candidates who were
described as handing out kilo bags of sugar, half-kilo bags of salt, and K500 bills to citizens in
exchange for votes at public rallies;16Candidates who promise citizens fertilizer subsidies, financial
aid for funerals, and help with other personal problems once elected in exchange for their votes;
and those who offer constituents to improve schools, improve healthcare, and dig more boreholes
once elected. In order to test the effect of negative campaigning concerning clientelism across
these three dimensions, we included three additional candidates who criticize each of these
platforms.17
Co-ethnicity and Local Ties. Although it is not the focus of this article, the experiment also includes
information on ethnicity and the strength of local ties. We include the nine largest ethnic groups,
which together make up 97 percent of the population of Malawi, in the randomization.18
Individuals who are members of these groups were presented randomly with their own ethnicity
about 50 percent of the time, and with one of the other ethnicities about 50 percent of the time.19
We also examine the impact of social ties by varying candidates’ backgrounds. Candidates were
described either as having grown up and remained in the village, signaling stronger community
16 The items chosen were informed by pre-survey focus group discussions (see Online Appendix E), and are also found in reports of the 2014 elections (MESN 2014). 17 The platforms of public goods provisions are informed by pre-survey focus group respondents (Online Appendix E) and qualitative interviews with Malawian MPs who in conversations explained what voters expected of their MP in terms of community development (see Rakner and Svåsand 2013). 18 See Berge et al. (2014, Table 1) for a complete listing of the ethnic groups and their percentages of the population. The nine largest Malawian ethnic groups we included in the experiment were: the Lomwe, the Yao, the Ngoni, the Tumbuka, the Mang’anja, the Sena, the Tonga, the Nkonde, and the Chewa. 19 The respondent’s ethnicity was based on an earlier question about his or her ethnicity. Anyone who is not a member of one of these nine largest ethnic groups was equally likely to see each of the possible ethnicity conditions, all of which were non-co-ethnics.
14
ties, or as having moved back to the village after time away. As noted above, each respondent saw
just one candidate profile. Full examples of potential candidate profiles are presented in Online
Appendix A.
Wealth Measures. Given the importance of poverty to our study, we measure respondent wealth in
four ways. It is important to note that we do not employ direct measures of income. Many
Malawians do not exist in a predominantly cash-based economy and, thus, respondents do not
always know their income. Instead, we employ alternative measures of wealth and class.
The first measure of wealth that we employ is an asset index. The asset index was created by
performing a multiple correspondence analysis on four assets a household could possibly possess:
motor vehicle, mobile telephone, radio, and bicycle. The higher the value, the more assets a
household possesses. This measure was standardized and cut into three categories based on asset
scores. Lower numbers represent respondents whose wealth according to an asset index is low,
higher scores represent scores closer to one on the asset index, meaning more assets and therefore
more wealth. According to this index, 41 percent of the respondents are in the most economically
disadvantaged bracket, 25 percent are in the middle wealth bracket, and 34 percent are in the most
economically advantaged bracket.
The second measure of wealth focuses on household income conditions. The question asks the
respondent which of the following statements is closest to their situation: their household income
covers their needs well and they can save, covers their needs without much difficulty, does not
cover their needs and there are difficulties, or does not cover their needs and there are great
difficulties. Only 81 respondents reported that their needs are covered and they can save. We
aggregated these very well-off individuals with the third highest category, those who are able to
cover their needs without much difficulty. We thus have three groups: households in which the
income covers needs without much difficulty (21 percent), households in which the income does
not cover needs and there are difficulties (33 percent), and households in which the income does
15
not cover the needs and there are great difficulties (46 percent).20
The third measure is based on a third-party perception of the respondent’s socioeconomic status.
Interviewers were asked to rate the status of the individual based upon the external appearance of
the house and neighborhood, compared to other houses in the same enumeration area. The
response options are: lower class (74 percent), middle class (24 percent), and upper class (3
percent). As there are only 32 respondents included in the experiment who were categorized in
the highest class category, these respondents were included in the upper class group. Thus, 27
percent of the sample is considered to be middle class and above.21
The fourth wealth measure taps into the housing infrastructure. This a standardized index created
from multiple component analysis of four questions that aim to get at the respondent’s standard
of living: the type of roof on the respondent’s home, the material the home is made of, and whether
the home has running water, and if it has electricity.22 The higher the score the better off the
respondent is thought to be doing based on the condition of their home. We cut this index into
thirds based on the scores, resulting in 15 percent of the respondents in highest wealth category,
35 percent in middle wealth category, and 50 percent in lowest wealth category.
20 See Online Appendix C for a disaggregated analysis of these income levels. The substantive results are robust to dividing this measure into high and low wealth halves. 21 The substantive findings of the experiment do not depend on this decision. 22 Homes with roofs made of thatch, grass, or plastic sheets are ranked as poorer households than those with roofs of metal, tiles, asbestos, or concrete. Homes made of burnt brick are rated as being better off than those of sun-dried brick, as well as those that are connected to the electrical grid or have solar power compared to those lacking electricity. Homes with tap water in the home or who relied on bottled mineral water as their source of water are ranked as better off than households that have a shared tap in a common place or a borehole, which are better off than those with a well or that rely on river, lake, or rain water. All variables are standardized by adding the absolute minimum to make the range positive and are divided by the absolute maximum plus the absolute minimum. This way all variables range from 0 to 1. x std = (abs(min(x))) + var (1) / (abs(max(x)) + abs(min(x))).
16
5. The Results
We used the four-point scale described above to rate candidates. We follow Hainmueller et al.
(2013) who show that Ordinary Least Squares (OLS) regression analysis is a consistent estimator
of the Average Marginal Component Effect (AMCE) of different candidate attributes on the
probability of a respondent voting for the candidate.23 To do so, we rescale the ratings to range
between 0 and 1.24 One level of each attribute is omitted to serve as the reference category.25 Here,
we present the impact of appeals, further conditioned on indicators of respondent’s wealth.
Average Marginal Component Effects: The Importance of Community Goods. The results demonstrate that
voters are not attracted to the promise of selective incentives. Using the promise of future
particularistic goods including fertilizer subsidies, financial aid for funerals, and help with other
personal problems once elected as the baseline appeal, OLS analysis finds that the ranking of
candidate platforms is as follows from the most to the least preferable platform: the long-term
promise of communal club goods, the long-term promise of individual benefits (access to services
or government benefits after the elections). Compared to the base of long-term selective
clientelism, immediate individually targeted goods (denoted as “Immediate Targeted Goods”
below) are significantly (p<0.001) less likely to be preferred by voters. (See Figure 1). Voters are
25 percentage points less likely to support candidates who attempt to buy their votes today than
they are to prefer those who offer targeted incentives in the future.
23 We do not run an analysis of the Complier Average Causal Effect (CACE) because we assume that we have all compliers or never compliers in this survey experiment. 24 We found no substantive differences in the estimated marginal effects analysis from an ordinal logit model or binary logit model that collapses those who replied “not at all likely” and “not likely” into a “candidate not preferred” group and those who replied “somewhat likely” and “very likely” into a “candidate preferred” group. All analyses were run with the weights and clustering developed for the GLD Malawi survey. We did not cluster at the individual level as each respondent only saw a single candidate profile. 25 Full results are presented in Online Appendix B.
17
Figure 1. Effects of Candidate Appeals (Base of Future Targeted Goods)26
Wealth and Support for Targeted Goods. To examine whether lower class individuals nevertheless prefer
candidates who offer selected incentives, we analyze heterogeneous effects in the conjoint
experiment across wealth levels. As shown in Figure 3, we find that the lowest wealth bracket, or
poorest Malawians, significantly prefers candidates who offer goods for the community by 23
percentage points and eschews immediate targeted goods (i.e., ‘vote buying’) by 28 percentage
points when compared to the promise of future selective goods. That is, voters appear to be driven
by community goods. The results are similar for those in the middle class. Respondents in the
middleincome group are 41 percentage points less likely to prefer a candidate who buys votes
outright when compared to those who offer future targeted benefits. However, in contrast to
assumptions underlying much of the literature, the upper wealth categories are not less likely to
26 These results hold up to the inclusion of controls for respondent characteristics including gender, age group, income level, education level, a measure of the respondent’s village etho-linguistic fractionalization score, and the region in which the respondent resides. These controls do not alter the results. Moreover, all substantive findings are robust to a binary “high” versus “low” rather than multi-level division of the wealth indicators. Since respondents saw a single profile only, responses are not clustered by respondent.
18
prefer particularistic goods. In short, there is no evidence that the poor are more likely to welcome
vote-buying than their wealthier co-nationals, nor that they are more likely to view co-ethnics more
favorably.
Figure 3. Effects of Candidate Appeals by Asset Index (Base of Future Targeted Goods)
Our confidence is increased by the fact that the findings are robust across the three
additional wealth measures: the respondent household’s ability to cover their needs, third-party
assessments of class, and a housing quality index. As shown in Figure 4, those who report that
their needs are not covered and are having great difficulties are statistically significantly less likely
to prefer candidates who offer to buy their votes by 30 percentage points compared to a candidate
who offers future targeted incentives, and significantly more likely to prefer candidates offering
community goods by 50 percentage points.27 The preference of those whose income needs are
well-met for those offering community goods is not statistically significant compared to a promise
of future particularistic goods. Figure 5 reports findings from the two categories of lower and
upper wealth categories as reported by enumerators. Individuals in the lower wealth category are
statistically significantly less likely to support candidates offering immediate, particularistic
27 Those in the middle wealth group, who have their needs met with some difficulties, are less likely to prefer candidates who offer immediate selective incentives by 24 percentage points compared to those offering future selective incentives.
19
incentives by 26 percentage points compared to those offering future targeted goods. By contrast,
those in the upper wealth category do not distinguish between offers of vote buying, future
clientelism, or club goods provision. Finally, using housing infrastructure as a measure of wealth
yields very similar results. (See Figure 6.) Those in the lowest wealth category by this measure are
statistically significantly more likely to prefer candidates who offer club goods (by 43 percentage
points) and future targeted goods (by 33 percentage points) to those who offer targeted handouts.28
Figures 4, 5, and 6. Effects of Candidate Appeals by Self-Reported Needs, by Third-
Party Assessment, and by Housing Infrastructure Index (Base of Future Targeted
Goods)
28 Somewhat surprisingly, those in the highest category of this index significantly preferred targeted goods (e.g., food stuffs and cash) handed out during the campaign to selective goods being handed out in the future by 21 percentage points.
20
In sum, our analyses find little support for the assumption, so prevalent in literature on vote-
buying, that the poor prefer clientelistic appeals. Regardless of the wealth indicator employed, we
do not find support for the hypotheses that individuals prefer candidates offering individual
incentives over those that offer community goods. We similarly find little support for the
hypotheses that Malawians are present-oriented, preferring targeted goods today to those in the
future. Rather, we find that the poor actually are less likely to support candidates who are willing to
buy votes. Vote buying not only fails to appeal to potential voters, it actually repels them.
6. Why Promises of Vote-Buying Repel Voters
Vote-buying is costly to candidates not only in terms of the goods they provide, but because the
act of vote buying alone repels poor voters. This is an important finding for scholars of clientelism,
but one may be concerned that the results of an experiment intended to focus on ‘pure’ or inherent
preferences does not have external validity. In this section, we consider the extent to which the
experimental results resonate in the ‘real world.’ We first explore insights from post-survey focus
groups, and then consider alternative explanations.
Insights from Focus Groups Participants in focus group discussions conducted after the survey underscored these dynamics.
We implemented twelve focus group discussions with separate groups of men, women and youth.
Nine of these were in lower income areas, and three in a middle class area.29 Focus group support
the experimental evidence that the poor often negatively view candidates who appeal to them
through vote-buying.
Our findings suggest that the poor were particularly likely to view candidates who focus on vote-
buying as untrustworthy. Both male and female focus groups in poor areas of Mzuzu and Dedza
29 The focus group discussion were carried out in September 2017. For information see Online Appendix E.
21
expressed the opinion that candidates who try to sway them by offers of money or goods are not
honest politicians and therefore do not deserve one’s vote.30 Similarly, most youth in Dedza
claimed that giving out money or items to voters is a sign of dishonesty, although some suggested
that it at least meant that such people help people at least during this campaign period. They too,
however, agreed that this was not a sign that the candidate would deliver.31 Experience has shown
that those who offer them goods do not visit their area anymore after winning.
Focus groups in the middle income area of Masasa also expressed distrust of candidates handing
out goods, but they did not appear to view them as hostilely. In part, this may be because they did
not view the candidates as treating them as inferiors, who can be bought, but rather, as equals.
Men in Masasa, for instance, said that they viewed the gifts as ‘compensation’ for their time.
Others suggested that the handouts were just to provide the fun.32 The middle income voters did
not view candidates offering gifts more positively than the poor, but they do not appear to have
been offended by them.
Participants in both lower and middle income groups generally agreed that they like receiving
handouts from politicians during the campaign period, but argued that their choices are not
affected by the gifts they pocket. One participant from a poor village in Dedza district explained,
“During campaign period it is our time to eat from politicians. Do we beg them?” (Nyengo yakampeni
nmdi nthawi yoti nafenso tiwadyere. Ngati tinawapempha)?()33 Another said, “When they bring the things
we receive and eat” (Munthu uja akabwera nazo timangolandira nkudya) (”). Most focus group
discussions expressed that respondents felt they had the freedom of choice.34
30 Focus group discussion, Malivenji village, Mzuzu, men, September 11, 2017; transcript from focus group (women), Kabinda, Dedza, September 12, 2017. 31Transcript from focus group (youth), Gunduze, Dedza, September 12, 2017. 32 Transcript focus group discussion (men), Masasa, Mzuzu, September. 11, 2017. 33 Transcript, focus group discussion (Women), Gunzuzi village, September 12, 2017. 34 Transcript, focus group discussion (Men), Kabinda village, Dedza, September 12, 2017.
22
Few say they felt that they were obligated to vote for a candidate who gave them money or gifts.35
One middle class respondent in Masasa explained that rewarding vote-buying was against civic
duty. He explained, “According to real politics we are not supposed to vote for somebody because
we have received something from them” (Kwakuyana na real politics, munthu tingamuvoteranga chifukwa
cha vinthu ivo watipasa yayi).36 A poor man in the Malenwenij village put it even more strongly,
focusing on the individual’s integrity. “But if we vote for them based on handouts these are the
ones who desert us wanting to recover his money” (Kweni pala tamuvotera chifukwa cha vila niwala
wakuoneka ha wali niwezge da ndalama zane), he explained. The extensive reliance of vote-buying
signals that the candidate is ill-intentioned.
Alternative Explanations We also explore alternative explanations, focusing on the potential validity of the experiment and
our interpretation. Before turning to this, it is important to note that the respondents held strong
convictions. Overall, 44 percent of the sample selected not at all likely to vote for the candidate
and 36 percent said that they were very likely to vote for the candidate.37 More than 75 percent of
respondents chose one of the extreme categories of either not at all likely or very likely to vote for
a candidate for each of the candidate appeals presented in this article.38
Despite this strong evidence, one may be concerned that the results are driven by social desirability
bias. Would people view reporting support for candidates who offer targeted goods as socially
undesirable, and thus be unlikely to do so? The conjoint experiment alleviates this problem to
35 These findings are corroborated by evidence from the LGPI. Only 22% of the entire survey sample reported that they had personally received gifts, food, or money from a candidate, and of these, 44% reported that they had received these handouts from multiple candidates. The nearly half of respondents reporting that they received handouts could not vote for more than one candidate. It is also consistent with evidence found elsewhere as well. For instance, an interlocutor in São Tomé explained, “We do like vote-buying. It is essential. That is the only way we have to see anything good coming from the politicians. Anyway, I can vote for whoever I want.” (Vicente 2014). 36 Transcript, focus group (men), Masasa, September 12, 2017. 37 The remaining 20 percent included 7 percent who were not likely to vote for the candidate, and 13 percent who were likely to do so. 38 The full distribution of results across appeals is available in Online Appendix B.
23
some extent, since avoiding direct questions regarding one’s willingness to vote for a candidate
offering selective goods diminishes the problems of social desirability bias. In addition, there is
evidence that Malawians do not distinguish between these offers as ones of clientelistic and
programmatic. In the LGPI, we included manipulation checks, asking respondents if they
perceived the offer they were given as ‘vote-buying.’ We find very little difference in their
perceptions of the three offers: about 64 percent of the respondents considered the short-term
clientelistic appeal to be vote buying; 63 percent viewed a particularistic, long-term clientelistic
appeal as vote buying; and 52 percent saw the offer of community goods as vote buying. The pre-
survey focus groups also corroborated these findings from the LGPI survey.39 Participants of the
pre-survey focus groups did not distinguish between immediate hand-outs and political promises
of “club goods” or national development post-election- they considered all as forms of vote-
buying.40 That is, a majority of Malawians consider each of the appeals presented in this paper to
be vote-buying. Thus, they should not be less likely to express enthusiasm for candidates offering
immediate, targeted goods than for those making other appeals. Nor would social desirability bias
explain why the poor are particularly likely to be repelled by these candidates. Social desirability
bias does not explain the results.
The experimental findings also do not appear to be driven by differences in the nature of the goods
presented in the experiment. One may be concerned that the immediate particularistic incentives
(e.g., cash and food stuffs in the election period) are less valuable than the future, selective
incentives (e.g., fertilizer subsidies, assistance with fees); thus, our findings that individuals prefer
future incentives over direct vote-buying may simply reflect that voters prefer more valuable
incentives over less valuable ones. We are unable to interrogate this fully. However, examining
39 See Online Appendix E. 40 Focus group discussions in Chiradzulu village, Southern region (men) provide an illustration. Participants listed as vote-buying the handing over of money, food, assistance for funerals and promise of fertilizer. It was generally considered that the incumbent had better means to vote-buying Those in power have money and campaigning machinery hence more powerful than those in opposition “ali m’boma angakaphwasule nkhokwe zonse za ku Lilongwe za chimanga pomwe wotsutsa alibe.” Meaning candidates whose party is in power can use government resources during their campaign while those in opposition cannot. Participant 7, transcript January 4, 2016).
24
heterogeneous effects by urban and rural populations (which presumably value fertilizer subsidies
more highly) finds that both populations prefer future goods to immediate goods.41 Moreover, this
does not contradict our most important finding: even the very poor prefer elected officials who
promise to aid their community over those who promise to help them individually, whether during
elections or in the future.
A related argument may be that in Malawi, everyone is poor, and thus, comparisons between
groups of different income levels are not relevant. The underlying concern is that we emphasize
relative poverty, while absolute poverty should also be important. However, as shown above, we
find that those whose needs are not met respond to vote-buying more negatively than those whose
needs are better met, which lends credence to our argument. Moreover our finding that those who
are less wealthy reject offers of immediate targeted goods holds if we compare only the individuals
within the top third of the asset index with those in the bottom third.42 They also hold across
disaggregation of each component of the asset index so that those who do not own a motor
vehicle, a mobile phone, a radio, or a bicycle are all significantly less likely to prefer this type of
good compared.43 For the housing index, while the results show that both the wealthy and the
poor reject vote buying looking at only those in the top third and bottom third of this index, when
disaggregated we see that the 67 percent of our sample who own homes made of burnt brick are
driving this finding; those without tap water or electricity at home as well as those who have roofs
made of weaker materials are all significantly more likely to reject targeted material handouts
compared to future targeted goods.44
41 About 93 percent of the sample live in rural areas farm land while only 60 percent of those in urban areas reported doing so. See Online Appendix D. 42 About 42 percent of the sample is in the low wealth category while 31 percent is in the high wealth category according to this breakdown. See Online Appendix C. 43 About 4 percent of the sample own a motorbike/vehicle, 54 percent own a mobile phone, 44 percent own a radio, and 35 percent own a bicycle. It is also notable that respondents in the upper most third of the housing index and those in the lowest third of this index are both statistically significantly less likely to prefer immediate targeted goods than future targeted ones. 44 See Online Appendix C.
25
Similarly, one might argue that it is not poverty, but lack of education, that drives the findings.
However, the extant literature would expect that it is the less educated (and poor) who are more
likely to accept vote-buying (e.g., Blaydes 2011).45 We find, however, that if we divide the sample
by education levels, those who have no primary school are not significantly less likely to prefer
candidates promising immediate targeted handouts, compared to those offering future targeted
goods. Rather those with some primary school education, which is 59 percent of our sample, drives
the rejection of vote buying. Those who have finished primary school and above are not
statistically significantly more or less likely to prefer candidates offering vote buying compared to
those offering future targeted goods.
A final concern may be that candidates “impute” offers of vote-buying, even to candidates that
offer other appeals. The assumption is that vote-buying is a minimum offer, and thus candidates
who were stated as making other appeals would be offering to buy votes as well. We have little
reason to believe this is the case. Vote buying definitely exists in Malawian elections, but it is not
ubiquitous. Only 22 percent of the entire survey sample in the 2016 LGPI reported that they had
personally received gifts, food, or money from a candidate, and similarly, the Varieties of
Democracy data suggests that vote buying in the 2014 elections was quite low with experts placing
it at 0.98 on a scale ranging from 0 to 4.46 We are unable to interrogate this concern fully, but it
does not appear that voters perceived all candidates as offering gifts, or that they view gift-giving
as a necessary offer to demonstrate their abilities.47 More importantly, our finding nevertheless
45 Note that Poverty and education are positively correlated, yet not identical (r=.37). Thus, we should not equate poverty with education. 46 According to the Varieties of Democracy Codebook, a 1 on this scale means, “There were non-systematic but rather common vote-buying efforts, even if only in some parts of the country or by one or a few parties” (Coppedge et al. 2017). 47 Indeed, our survey finds evidence in contrast to arguments that individuals may view vote-buying as reflecting a candidate’s ability to provide for the community. Kramon (2016) argues, for instance, that a preference for candidates offering selective incentives may be understood as a preference for candidates signaling capacity and commitment to the community. However, Malawians’ responses to the direct question regarding whether promising targeted goods today or in the future draws this into question. We find that more than 80 percent of Malawians do not view the provision of such goods as “demonstrating the ability of the candidate to provide for the community.”
26
importantly points to the fact that vote-buying in the absence of other offers is viewed as offensive,
particularly by poor voters.
7. Conclusion
The findings in this study draw into question widespread assumptions underlying the literature on
clientelism. Scholars of vote-buying have largely assumed that vote selling candidates target the
poor, and the poor grant them support in return. We find, however, that poor voters are less likely
to support candidates who offer them immediate targeted incentives, and they prefer those who
promise to deliver community goods. The poor are willing to accept offers of cash, sugar and other
handouts at election time, but they question their motives, seeing them as more interested in
winning elections than in the welfare of the community. Citizens may accept material handouts at
election time, but they do not view themselves as committed to voting for the candidate just
because they do so. Vote-buying candidates lose support of the poor when they are seen as
monetizing the vote.
This raises important questions regarding why, and when, vote-buying increases vote shares for
these candidates. Certainly, vote selling is not always negatively correlated with vote share, as one
might expect if vote selling candidates repel voters. But, when is it effective? The distribution of
goods and services may be effective if it is part of long-term, clientelistic relationships. In this case,
handouts do not create credibility (as Kramon 2016 suggests) but rather are part of continued
exchange. They are inoffensive only where the candidate is credible from the outset. Vote-buying
may also be effective when paired with appeals of community. In this case, voters can enjoy the
fruits of election while justifying their vote choice in terms of community welfare. The exchange
is not viewed as one of cash for votes (i.e. voting is not monetized), and the exchange is not
offensive.
27
More research should be done to understand when vote-selling offers repel or attract votes. We
should extend the focus beyond one on how income levels influence demand for vote-selling (e.g.,
Weitz-Shapiro), to one that more clearly takes into account the social norms and institutions that
shape attitudes toward and responsiveness to vote-sellers. We also should take into account more
clearly the different nature of vote-sellers -- that is, whether offers are made directly by candidates
or by intermediaries, and by those with whom voters have or do not have a personal connection.
That is, we should consider that the mechanisms at work go beyond the immediate exchange of
money for votes.
The study also suggests the need for more research on the preferences of the poor. Studies on the
poor primarily drawn from social psychology and implemented in the West have emphasized the
prosocial attitudes of the poor, while those drawn primarily from behavioral economics and
political science, more frequently implemented in the Global South, have focused on the poor’s
tendency toward time discounting. These depictions are not necessarily at odds with each other -
- even time-discounting poor may hold prosocial attitudes. But, more work needs to be done to
examine the relationship between these findings and to consider the extent to which these findings
hold across very different contexts.
Future work on the preferences of the poor concerning clientelism should explore the extent to
which the findings set forth here transfer to settings that vary in terms of culture, economic
development, inequality, and electoral conditions. In this study, we have focused on parliamentary
elections elected in first-past-the-post, single member districts in a largely poor, underdeveloped
country. The findings are important, as they demonstrate that even those often teetering on the
brink of survival are highly critical of clientelism. Even in the ‘hard case’ of Malawi, we find
evidence against the widespread assumption that the poor embrace clientelism. Yet much work
remains if we are to understand how this varies across conditions, and the mechanisms at work.
28
References
Auyero, J. (1999). From the client’s point(s) of view: How poor people perceive and evaluate
political clientelism. Theory and Society, 28: 297-334.
Auyero, J. (2000). The Logic of Clientelism in Argentina: An Ethnographic Account” Latin
American Research Review, 35(3): 55-81.
Ballington, J., & Kahane, M. (2014). Women in Politics: Financing for Gender Equality. In E.
Falguera, S. Jones & M. Ohman (Eds.), Funding of Political Parties and Election Campaigns.
A Handbook on Political Finance, 301-343. Stockholm: International IDEA.
Banful, A. (2011). Old problems in the new solutions? Politically motivated allocation of
program benefits and the new fertilizer subsidies, World Development, 39(7): 1166-1176.
Berge, E., D. Kambewa, Munthali, A., and Wiig, H. (2014). Lineage and land reforms in Malawi:
Do matrilineal and patrilineal landholding systems represent a problem for land reforms
in Malawi? Land Use Policy, 41: 61-19.
Bjørkman, L. (2013). You can’t buy a vote: cash and community in a Mumbai election. MMG
Working Paper, (13-01).
Blaydes, L. (2011). Elections and Distributive Politics in Mubarak’s Egypt, New York: Cambridge
University Press.
Bleck, J., & Van de Walle, N. (2011). Parties and issues in Francophone West Africa: towards a
theory of non-mobilization. Democratization, 18(5): 1125-1145.
Bowles, S. (2016). Incentives Are No Substitute for Good Citizens. New Haven: Yale University Press.
Carlson, E. (2015). Ethnic Voting and Accountability in Africa: A Choice Experiment in
Uganda, World Politics, 67: 353-385.
Carvalho, L., Meier, S., and Wang, S. (2016). Poverty and Economic Decision-Making:
Evidence from Changes in Financial Resources at Payday. American Economic Review,
29
106(2): 260-84.
Cox, W. and McCubbins, M. (1986). Electoral Politics as a Redistributive Game, The Journal of
Politics, 48: 370-389.
Coppedge, M., Gerring, J., Lindberg, S.I., Skaaning, S.E., Teorell, J., Altman, D., Bernhard, M.,
Fish, M.S., Glynn, A., Hicken, A. and Knutsen, C.H. (2017). V-Dem Dataset v7, V-Dem
Reference Material.
Corstange, D. (2016). The Price of a Vote in the Middle East: Clientelism and Communcal Politics in
Lebanon and Yemen. Cambridge University Press.
Corstange, D. (2017) “Clientelism in Competitive and Uncompetitive Elections”, Comparative
Political Studies, 1-29.
Desposato, S. (2007). How does vote buying shape the legislative arena. Elections for sale: The
causes and consequences of vote buying, Boulder, CO: Lynne Rienner: 101-122.
Diaz-Cayeros, A., Magaloni, B., and Estevez, F. (2016). The political logic of poverty relief: Electoral
strategies and social policy in Mexico, Cambridge University Press.
Dietze, P. and Knowles, E. (2016). Social Class and the Motivational Relevance of Other Human
Beings, Psychological Science. 27(11): 1517-152.
Dionne, K. and Dulani, B. (2013). Constitutional Provisions and Executive Succession: Malawi’s
2012 Transition in Comparative Perspective, African Affairs, 112(446): 111-137.
Dixit, A., & Londregan, J. (1996). The determinants of success of special interests in
redistributive politics. the Journal of Politics, 58(4), 1132-1155.
Dulani, B., & van Donge, J. (2005). A Decade of Legislature-Executive Squabble in Malawi,
1994–2004. African Parliaments: 201-224. Palgrave Macmillan US.
Ejdemyr, S., Kramon, E., and Robinson, A. (2015) “Segregation, Ethnic Favoritism, and the
Strategic Targeting of Local Public Goods.” Presented at the American Political Science
30
Association Annual Conference.
Ferree, K. (2010). Framing the Race in South Africa: The Political Origins of Racial Census Elections.
Cambridge University Press.
Finan, F. and Schechter, L. (2012). Vote-Buying and Reciprocity, Econometrica, 80: 863–881.
Franchino, F. and Zucchini, F. (2015) Voting in a Multi-dimensional Space: A Conjoint Analysis
Employing Valence and Ideology Attributes of Candidates, Political Science Research and
Methods, 3(2): 221–241.
Gonzalez-Ocantos, De Jonge, C. K., Meléndez, C., Osorio, J., and Nickerson, D. W. (2012).
“Vote Buying and Social Desirability Bias: Experimental Evidence from Nicaragua.
American Journal of Political Science, 56: 202-217.
Gonzalez-Ocantos, E. de Jonge, C. K., Meléndez, C., Nickerson, D., and Osario, D. (2013).
“Carrots and Sticks: Experimental Evidence of Vote Buying and Voter Intimidation in
Guatemala”, Annual Meeting of the APSA, Chicago: August.
Gonzalez Ocantos, E., de Jonge, C. K. and Nickerson, D. W. (2014), The Conditionality of
Vote Buying Norms: Experimental Evidence from Latin America. American Journal of
Political Science, 58: 197–211.
Grossman, G. M. and Helpman. E. (1996). Electoral Competition and Special Interest Politics,
The Review of Economic Studies, 63: 265-286.
Hainmueller, J., Hopkins, D. J., & Yamamoto, T. (2013). Causal Inference in Conjoint Analysis:
Understanding Multidimensional Choices via Stated Preference Experiments, Political
Analysis, 22: 1-30.
Hainmueller, J., Hangartner, D. and Yamamoto, T. (2015). Validating vignette and conjoint
survey experiments against real-world behavior. Proceedings of the National Academy of
Sciences, 112(8): 2395-2400.
Hainmueller, J. and Hopkins, D. (2015). The Hidden American Immigration Consensus: A
31
Conjoint Analysis of Attitudes toward Immigrants, American Journal of Political Science, 59:
495-510.
Haushofer, J., and Fehr, E. (2014). On the psychology of poverty. Science, 344(6186): 862-867.
Hicken, A. (2011). Clientelism, Annual Review of Political Science,14: 289-310.
Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the world? Behav
Brain Sci. 33(2-3), 61-83.
Jensen, P. and Justesen, M. (2014). Poverty and vote buying: Survey-based evidence from
Africa. Electoral Studies 33: 220–232.
Justesen, M. and Manzetti, L. (2017). Poverty, Partisanship and Vote Buying. Unpublished
manuscript.
Kaspin, D. (1995). The politics of ethnicity in Malawi´s democratic transition. The Journal of
Modern African Studies, 33(4): 595-620.
Kitschelt, H. (2000). Linkages between Citizens and Politicians in Democratic Politics,
Comparative Political Studies, 33: 845-879.
Kitschelt, H. (2002). Popular Dissatisfaction with Democracy: Populism and Party Systems,
Democracies and the Populist Challenge, Palgrave Macmillan, London.
Kramon, E. (2016). Electoral handouts as information: Explaining unmonitored vote buying,
World Politics, 68: 454-498.
Kraus, M. W., Piff, P. K., & Keltner, D. (2009). Social class, sense of control, and social
explanation, Journal of Personality and Social Psychology, 97(6): 992-1004.
Lawrance, E. (1991). Poverty and the rate of time preference: evidence from panel data. Journal
of Political Economy, 99(1): 54-77
Lindberg, S. I., & Morrison, M. K. (2008). Are African voters really ethnic or clientelistic? Survey
32
evidence from Ghana, Political Science Quarterly, 123(1): 95-122.
Lust, E., A. Harris, K. Kao, P. Landry, B. Dulani, A. Zgambo, A. Chiweza, H. Kayuni, R.
Muriaas, L. Rakner, V. Wang, L. Benstead, F. Hartmann, and S. Nickel. (2016). The
Local Governance Performance Index (LGPI) Malawi Truncated Dataset. Program on
Governance and Local Development, University of Gothenburg.
Magaloni, B. (2006). Voting for Autocracy. New York: Cambridge University Press
Malawi Electoral Support Network. (2014). Election Situation Room Draft Report. EU Malawi. Final
Report. Tripartite elections. Presidential, parliamentary and local council. 20 May 2014.
Brussels: European Union Election Observation Mission.
Mares, I. and Young, L. (2016). Buying, Expropriating and Stealing Votes. Annual Review of
Political Science, 19: 267-288.
Mpesi, A.M. and Muriaas, R. (2012). Food security as a political issue. The 2009 elections in
Malawi. Journal of Contemporary African Studies, 30(3).
Munzo, P. (2014). An informational theory of campaign clientelism: The case of Peru.
Comparative Politics, (47)1: 79-98.
The Nation. (2013). Eight Years of CDF: Any Hope? Oct 2013. Available at:
http://www.mwnation.com/eight-years-of-cdf-any-hope/
Nyasa Times. (2014). Kaliati Defends Handouts to Voters. Jan 2014. Available at:
http://www.nyasatimes.com/kaliati-defends-handouts-to-voters/
Nichter, S. (2008). Vote Buying or Turnout Buying? Machine Politics and the Secret Ballot,
American Political Science Review, 102: 19-31.
Patel, N. and Wahman, M. (2015). The Malawi 2014 Tripartite Elections: Is Democracy
Maturing? Lilongwe: National Institute of CivicEducation (NICE).
Pellicer, M., Piraino, P., & Wegner, E. (2016). Is Inequality Inevitable? A Survey Experiment on
33
Demand for Redistribution in South Africa. German Institute of Global and Area
Studies. Unpublished manuscript.
Piff, P., Kraus, M., Côté, S., Cheng, B., and Keltner, D. (2010). Having less, giving more: The
influence of social class on prosocial behavior. Journal of Personality and Social Psychology,
99(5): 771-784.
Piff, P., Kraus and Robinson, Amanda R. (2017). Social class and prosocial behavior: current
evidence, caveats, and questions. Current Opinion in Psychology.
Posner, D. (2004). The political salience of cultural difference: Why Chewas and Tumbuka are
allies in Zambia and adversaries in Malawi. American Political Science Review 98(4): 529-545.
Rakner, L. and Svåsand, L. (2013). Competition and participation, but no democracy: The
politics of elections in Africa´s feckless regimes. Statsvetenskapelig Tidskrift, 115(4): 365-
384.
Scott, J. (1969). Corruption, Machine Politics, and Political Change, The American political science
review 63(4): 1142-1158.
Shah, A., Mullainathan, S., and Shafir, E. (2012). Some Consequences of Having Too Little.
Science, 338: 682-685.
Stellar, J.E., Manzo, V., Kraus, M., and Keltner, D. (2012). Class and Compassion:
Socioeconomic Factors Predict Responses to Suffering, Emotion, 12(3): 1-11.
Stephens, N., Markus, H., and Townsend, S. (2007). Choice as an Act of Meaning: The Case of
Social Class. Journal of Personality and Social Psychology, 93(5): 814 – 830.
Stokes, C. S. (2005). Perverse Accountability: A Formal Model of Machine Politics with
Evidence from Argentina, American Political Science Review, 99: 315-325.
Stokes, C. S., Dunning, T., Nazareno, M., and Brusco, V. (2013), Brokers, Voters, and Clientelism:
The puzzle of distributive politics, Cambridge University Press.
34
Szwarcberg, M. (2012). Uncertainty, Political Clientelism and Voter Turnout in Latin America:
Why Parties Conduct Rallies in Argentina. Comparative Politics 45(1): 86–106.
Szwarcberg, M. (2015). Mobilizing Poor Voters: Machine Politics, Clientelism, and Social Networks in
Argentina. Cambridge University Press.
Tambulasi, R. and Kayuni, H. (2002). Can African feet divorce Western shoes? The case of
Ubuntu and democratic good governance in Malawi, Nordic Journal of African Studies,
(14)2: 147-161.
Tanaka, T., Camerer, C., and Nguyen, Q. (2016). Risk and time preferences: linking experimental
and household survey data from Vietnam. Behavioral Economics of Preferences, Choices, and
Happiness. 3-25.
Vicente, P. C. (2014) Is Vote Buying Effective? Evidence from a Field Experiment in West
Africa, The Economic Journal, 124: 356-387.
Weitz-Shapiro, R. (2014). Curbing Clientelism in Argentina: Politics, Poverty, and Social Policy.
Cambridge University Press.
World Bank. (2017). GDP Per Capita, World Bank National Accounts Data. Available at:
http://data.worldbank.org/indicator/NY.GDP.PCAP.CD?year_high_desc=false
35
Online Appendix A: Experimental Design
I am about to read you the descriptions of a candidate for parliament. Then I will ask you how
likely you would be to vote for this parliamentary candidate.
Table OA1: Some Example Candidates (The respondent saw only one candidate.)
Candidate
Example 1
A local resident (of respondent’s ethnicity) who was born in (this village/ward
input from earlier question) and has lived in the area for a long time is running as
a candidate for the parliamentary elections. At public rallies he hands out kilo
bags of sugar, half-kilo bags of salt, and K500 bills to citizens in exchange for
votes.
Candidate
Example 2
A man (of a non-co-ethnic ethnicity) who has recently moved to back to (this
village/ward input from earlier question) after many successful years living
abroad is running as a candidate for the parliamentary elections. At public rallies,
he offers citizens fertilizer subsidies, financial aid for funerals, and help with other
personal problems once elected in exchange for their votes.
Candidate
Example 3
A local resident (of a non-co-ethnic ethnicity) who was born in (this village/ward
input from earlier question) and who has lived in the area for a long time is
running as a candidate for the parliamentary elections. At public rallies, he
emphasizes to citizens that he will pass legislation to improve schools, improve
healthcare, and dig more boreholes once elected in exchange for their votes.
The interviewer reads the description and asks the respondent: “How likely is it that you would
vote for this parliamentary candidate: very likely, somewhat likely, not likely, not at all likely. Or
would you like me to read the description of the candidate again?” When the respondent indicates
that he or she is ready to answer, the interviewer records the answer.
36
Online Appendix B: Additional Information on the Experimental
Outcomes
Table OA2. Determinants of Preferences for Candidate Attributes Without Respondent
Controls (Base is Targeted Future Goods)48 in OLS,49 logit, and ordinal logit with 3 or 4
categories.
Candidate Attribute OLS Logit Ologit 3 Ologit 4
Local 0.075** 0.353* 0.325** 0.324**
(0.026) (0.135) (0.111) (0.111)
Co-ethnic 0.006 -0.0438 0.0618 0.0484
(0.036) (0.164) (0.159) (0.157)
Community Goods 0.113 0.483 0.494 0.494
(0.069) (0.345) (0.287) (0.289)
Immediate Targeted Goods -0.269*** -1.222*** -1.181*** -1.186***
(0.050) (0.257) (0.197) (0.201)
Immediate Targeted Goods Negative -0.051 -0.228 -0.215 -0.210
48 Survey weights were used in all analyses because the aim of this estimation is to test for the existence of effects within the general population of Malawian voters. 49 The OLS outcome is rescaled to range between 0 and 1.
37
(0.064) (0.307) (0.256) (0.263)
Future Targeted Goods Negative -0.047 -0.261 -0.159 -0.178
(0.051) (0.239) (0.203) (0.203)
Community Goods Negative -0.196** -0.835* -0.826** -0.825**
(0.059) (0.297) (0.242) (0.243)
Constant 0.501*** 0.131
(0.045) (0.214)
Cut 1
-0.355 -0.366
(0.188) (0.190)
Cut 2
0.503* -0.0743
(0.188) (0.177)
Cut 3
0.492*
(0.186)
Observations 1165 1165 1165 1165
Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001
38
Figure OA1. OLS Determinants of Preferences for Candidate Atributes Full Model (Base
of Future Goods)
Table 2. Determinants of Preferences for Candidate Attributes With Respondent Controls
(Base is Targeted Future Goods for Appeals) in OLS, logit, and ordinal logit with 3 or 4
levels.
Candidate Attribute OLS Logit Ologit 3 Ologit 4
Local 0.070* 0.363* 0.322* 0.321*
(0.027) (0.140) (0.120) (0.121)
Co-ethnic 0.167 0.007 0.126 0.109
39
(0.038) (0.187) (0.181) (0.179)
Community Goods 0.137* 0.617 0.600* 0.604*
(0.063) (0.321) (0.256) (0.258)
Immediate Targeted Goods -0.254*** -1.204*** -1.172*** -1.173***
(0.052) (0.275) (0.222) (0.223)
Immediate Targeted Goods Negative -0.023 -0.103 -0.104 -0.0988
(0.059) (0.297) (0.254) (0.259)
Future Targeted Goods Negative -0.004 -0.0486 0.0419 0.0240
(0.051) (0.247) (0.205) (0.203)
Community Goods Negative -0.191** -0.864** -0.803** -0.799**
(0.054) (0.277) (0.248) (0.247)
Female -0.283 -0.200 -0.0596 -0.0797
(0.028) (0.146) (0.128) (0.127)
Above 25 and Below 35 0.009 0.0988 0.0300 0.0328
(0.040) (0.196) (0.193) (0.188)
Above 35 and Below 45 0.005 0.0646 0.0185 0.0175
(0.049) (0.286) (0.236) (0.237)
Above 45 and Older 0.001 0.0279 -0.0216 -0.0225
40
(0.038) (0.199) (0.196) (0.190)
Some Primary Schooling -0.145** -0.726** -0.701** -0.712**
(0.058) (0.206) (0.211) (0.210)
Primary School Completed -0.263*** -1.279*** -1.298*** -1.309***
(0.058) (0.295) (0.289) (0.287)
Intermediate to Postgraduate -0.170* -0.957** -0.746* -0.785*
(0.065) (0.301) (0.309) (0.306)
ELF Quartile 2 -0.008 0.00919 -0.0391 -0.0407
(0.025) (0.119) (0.136) (0.125)
ELF Quartile 3 -0.001 0.0577 0.0165 0.0186
(0.032) (0.151) (0.161) (0.156)
ELF Quartile 4 0.014 0.213 0.000740 0.0210
(0.049) (0.267) (0.243) (0.242)
Northern Region 0.239*** 1.267*** 0.942*** 0.962***
(0.048) (0.271) (0.197) (0.197)
Southern Region 0.008 -0.0109 0.0300 0.0193
(0.034) (0.173) (0.164) (0.160)
Constant 0.587*** 0.488
(0.087) (0.444)
41
Cut 1 -0.725 -0.758
(0.407) (0.402)
Cut 2 0.186 -0.445
(0.416) (0.391)
Cut 3 0.156
(0.408)
Observations 1149 1149 1149 1149
Standard errors in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001
Table 3. Distribution of candidate rating in the experiment: Frequency Percent
Not at all likely to vote for the candidate 463 44.2
Not likely to vote for the candidate 92 6.7
Likely to vote for the candidate 174 12.8
Very likely to vote for the candidate 436 36.3
Total 1,165 100
Table 3. Distribution of candidate rating in the experiment:
Distribution of candidate rating in two groups with the two lower
categories collapsed into 0 and the two higher categories collapsed
into 1:
Frequency Percent
0 – Candidate is not preferred 555 50.9
1 – Candidate is preferred 610 49.1
Total 1,165 100
Distribution of candidate rating in three groups with the two middle
categories collapsed into a middle category:
Frequency Percent
42
0 – Candidate is not preferred 463 44.2
1 – Candidate is neither preferred or not preferred 266 19.5
2 – Candidate is preferred 436 36.3
Total 1,165 100
Distribution of candidate rating
by appeal:
Not at all
likely to vote
for the
candidate
Not likely to
vote for the
candidate
Likely to vote
for the
candidate
Very likely
to vote for
the
candidate
Immediate Targeted Goods 65.4% 6.5% 9.7% 18.3%
Future Targeted Goods 36.2% 7.0% 15.9% 41.0%
Community Goods 26.8% 4.9% 13.3% 55.0%
Total
Distribution of candidate appeals
including negative campaigning:
Number of Respondents Proportion
1 - Immediate Targeted Goods 189 .1681
2 - Future Targeted Goods 194 .1535
3 – Negative Campaign Immediate
Targeted Goods
202 .1696
4 – Negative Campaign Future Targeted
Goods
218 .1782
5 – Negative Campaign Club Goods 162 .1377
6 - Club Goods 224 .1929
Total 1189 1
*The sample is weighted so the numbers and percentages do not always directly correspond to one another.
43
Distribution of just the three positive
appeals:
Number of Observations Proportion
Immediate Targeted Goods 189 .3159
Future Targeted Goods 194 .3008
Club Goods 224 .3833
Total 607 1 *The sample is weighted so the numbers and percentages do not always directly correspond to one another.
Online Appendix C. Distribution and Question Wording of Wealth
Indicators
I will read out a few statements about your income.
Please tell me, which of the following statement is
closest to your situation?
Number of
Respondents
Percent
Our household income covers the needs well - we can
save.
516 8.4
Our household income covers the needs alright,
without much difficulty.
838 12.8
Our household income does not cover the
needs, there are difficulties.
2692 32.6
Our household income does not cover the
needs, there are great difficulties.
3580 46.0
Don’t know/Refuse 24 0.4
Total 7651 100 *The sample is weighted so the numbers and percentages do not always directly correspond to one another.
44
What is the socioeconomic status of the housing
based upon the external appearance of the
house and neighborhood compared to other
houses in the same EA?
Number of Respondents Percent
Lower class 4082 53.9
Middle class 2410 30.0
Upper class 1084 16.1
Total 7576 100
*The sample is weighted so the numbers and percentages do not always directly correspond to one another.
Income Measure #1: Asset Index:
Asset Index Components Number Yes – Own Item Percent
Motorbike/vehicle 45 3.7
Mobile phone 654 54.4
Radio 488 43.8
Bicycle 411 34.5
Total 7667 100
Distribution of asset index Number of Respondents Percent
1 3249 42.4
2 1995 26.0
3 2423 31.6
Total 7667 100
*The sample is weighted so the numbers and percentages do not always directly correspond to one another.
45
46
Income Measure #2: Self-Reported Needs:
Income Measure # 4: Housing Index, Class Coded by Enumerator Based on How the
House Looks from the Outside
Housing Number of Observations Percent
1 4158 51.5
2 3083 38.2
3 833 10.3
Total 8065 100
47
Housing Index cut into halves
48
Online Appendix D. Other Robustness Checks
*The sample is weighted so the numbers and percentages do not always directly correspond to one another.
Ethnolinguistic
Fractionalization
Index
No. of
Respondents in
dataset
Percent in
dataset
Number of
Respondents in
Experiment
Percent in
Experiment
Low Diversity 2180 26.6 303 26
Middle-low
Diversity
2231 25.9 275 22.4
Middle-high
Diversity
2350 30.1 306 25.3
High Diversity 1186 17.3 269 26.3
Total 7947 100 1153 100 *The sample is weighted so the numbers and percentages do not always directly correspond to one another.
Education
If we divide the sample by education levels we see that those who have no primary school are not
significantly less likely to prefer immediate targeted handouts compared to future targeted goods.
However, those with some primary school education, which is 59% of our sample, drives the
rejection of vote buying finding to a statistically significant degree. Having finished primary school
as well as an education level of intermediate school and above do not have a significant effect on
preference for a candidate offering vote buying compared to one offering future targeted goods.
49
Ethnolinguistic Fractionalization: Populations living in areas within the two lowest ELF
quartiles are the least likely to prefer targeted handouts and to a significant degree compared to
future selective benefits. This indicates there is a relationship beteween ethnic homogeneity and
dislike of vote buying. Only the least diverse areas are significantly more likely to prefer commnuity
goods than other types of goods.
Urban and Rural Subdivision: Analysis of the results by an urban-rural division reveals that that
rural residents prefer immediate targeted goods to a significant degree more than future selective
beneifts.
50
Regional Subdivision: Analysis of the results by an regional division reveals that that Northern
residents are significantly different in their candidate preferences compared to Central or Southern
residents. The Southern region is significantly likely to reject offers of direct vote buying whereas
inhabitants of the Central or Northern regions are not.
51
Party Last Voted For in the Parliamentary Elections: DPP and MCP supporters are
significantly likely to reject candidate who offer targeted handouts compared to other parties.
These parties took first and third in the elections.
Table 4. Views toward Vote Buying and Elections
Sometimes candidates offer voters access to
services and benefits (such as offering
voters fertilizer subsidies, financial aid to
pay funeral fees, and/or other help with
personal problems) once they are elected.
Would you say that doing this is ...
Sometimes candidates
offer gifts, food, or
money to voters during
the campaign season
would you say that doing
this is ...
Corrupting the
democratic process
49% No
46% Yes
4% Don't Know/Refuse
47% No
49% Yes
5% Don't Know/Refuse
52
The only thing one
can get out of
elections
65% No
32% Yes
4% Don't Know/Refuse
66% No
31% Yes
4% Don't Know/Refuse
Demonstrating the
ability of the
candidate to provide
for the community
82% No
16% Yes
3% Don't Know/Refuse
85% No
12% Yes
3% Don't Know/Refuse
The figure below shows differences among villages in which above 50% of the inhabitants report
that candidates in their communities offer handouts of gifts, money, or promise access to
government services, versus below 50% of village inhabitants reporting that this occurs during
elections (villages at 50% were dropped). Villages with high reports of vote buying are
significantly more likely to reject immediate targeted goods than future ones or community
goods. Villags with low reports of vote buying are also less likley to prefer immediate targeted
handouts from candidates than future ones, and they are also significantly more likely to prefer
community goods.
53
Online Appendix E: Survey Methodology
We conducted two rounds of focus group discussions: One prior to implementing the survey,
January 2-4 2016 (Table 1), and one after our initial analysis of the survey data, September 2017
(Table 2). The focus group discussion (FGD) participants at each site were screened by age and
gender. Following standard FGD protocols, three groups were convened at each site, one made
up entirely of women and one of men one of youth (gender mixed). The discussion checklist was
designed to elicit in-depth discussion on the study objectives, while being mindful that the basic
operative rule for FGDs requires that facilitation should, as much as possible, be kept to a
minimum.
In order to ensure open and full discussions, the fieldworkers were advised to limit the number
of FGD participants to a minimum of six and a maximum of ten. This was done in conformity
with global FGD guidelines that suggest that large groups may often result in a higher number of
participants who remain uncommunicative. The discussions were conducted in Tumbuku and
Chichewa, by two Malawian enumerators in each group.
54
Pre-LGPI Malawi Focus Groups: January 2016
Before implementing the LGPI survey, we conducted 18 focus groups, for men, women and
youth in the three regions of Malawi: Northern, Central and Southern. In each region we
selected one urban and one rural district for the focus group discussions. We also used these
focus groups to determine the types of goods candidates are most likely to promise during the
campaign season. A total of 174 people participated in the focus group discussions that were
carried out between January 4 and January 6, 2016.
The main purpose of the focus group was to ensure a common understanding and use of terms
and concepts. The focus group discussions were structured into four main themes: Key terms,
the role and functions of various local authorities, service delivery, power, politics and voting.
Pre-Survey Focus Group Discussions Village/area of
discussion
Number of
focus group
participants
(youth)
Number of
focus group
participants
(male)
Number of
focus group
participants
(female)
Northern region
Urban Mzuzu City
(Nkhata Bay)
10 11 13
Rural Timbiri
(Nkhata Bay)
10 11 8
Southern region
Urban Kachere 10 11 10
Rural Chiraduzulu 7 14 10
Central region
Urban Lilongwe Area 25 11 8 13
Rural Ntcheu District 9 10 8
TOTAL 57 65 52
55
The Malawi LGPI Survey
Sampling and Weighting
The Local Governance Performance Index (LGPI) survey was conducted in Malawi during
March and April 2016. This survey seeks to measure and better understand governance and
service delivery at the local level. Importantly, this is a highly clustered survey, which facilitates
measurement and inference at the local (in this case, village) level. The survey covers the
following topics: political participation, social norms and institutions, education, health, security,
welfare, corruption, land, and dispute resolution.
The sample was stratified on region (North, Central, South), the presence of matrilineal and
patrilineal ethnic groups, and the ‘urban’/rural divide. Because patrilineal groups are rare in
Malawi and we wanted to maximize variation in matrilineal and patrilineal heritage, we
oversampled Primary Sampling Units (PSUs) from the patrilineal stratum. We sampled 22 PSUs,
namely ‘Traditional Authorities’ (TAs). These 22 sampled TAs are located in 15 of Malawi’s 28
districts.50 Within each TA (i.e., PSU), we selected randomly four enumeration areas (EAs) as
Secondary Sampling Units (SSUs). EAs are comparable to census tracts. Both PSUs and SSUs
were selected without replacement according to the principle of Probability of Selection
Proportional to Measure of Size (PPMS). Within each EA, we sampled four villages, based on
known geographical points provided on the maps of the EAs produced for Malawi’s latest
population census. Once in the village, enumerators followed a random walk pattern to select
households. After they entered the household, the interviewer collected the necessary data about
composition of the household . Both the contact questionnaire and the main questionnaire we
programmed on digital tablets, including the selection of the final respondent in the household
through a digital version of the “Kish grid”. The target was to interview 22 respondents in each
village. This process produced a sample of 8,100 respondents. See Table 1 for a list of the
districts and TAs included in the sample and Table 2 for a list of the villages.
While the sampling procedures were planned as presented, of course in practice this was not
always the case. In total the research team had to draw 11 replacement EAs. One replacement
50 Districts are the largest sub-national administrative units in Malawi.
56
EA was drawn because enumerators were chased out of a village and forced to withdraw from
the EA. In the remaining 10 cases, EAs were not accessible (e.g. in one instance our team was
unable to reach the designated EA because a bridge had washed away during heavy rains.). In
these instances, backup enumeration areas were randomly selected within the same EAs
(excluding already selected and inaccessible zone) and were used as replacements. Such cases, the
variable “ea_replacement” in the data takes a value of 1.51 In addition, given that multiple
enumerators conducted surveys in the same village, the target number of 22 respondents per
village (neighborhood in urban areas) was not always reached precisely. In some instances more
were surveyed and in others slightly fewer than 22 households were surveyed. In addition, the
boundaries between villages and neighborhoods were not always clear, which also caused our
teams to deviate from the target of 22 per village/neighborhood.
In order to weight the sample for analysis, we constructed national-level post-stratification
weights based on the 2008 census data based district population size, education, gender,
ethnicity, and age.52 This ensures that the analysis is representative of the Malawian population.
In addition, the probability weights and finite population correction factors were computed at
the PSU and SSU level. We were unable to obtain valid village level population estimate because
villages are not included in the Malawian statistical system, in large part due to frequent
fluctuations in village leadership and boundaries. Thus, the probability weight computation
implies that households were drawn from SSUs directly and ignores the screening process at the
village level. In short, the weights provided in the dataset are applicable to the TA and EA levels,
but not the village leve.
51 In total, only 11 of the 99 sampled EAs are replacement EAs. 52 While we do use census data to create the weights, it is data obtained from IPUMS (Integrated Public Use Microdata Series) and is a random sample of 10% of the population. In the dataset, the variable “district_pop” contains the population size of the district. In STATA, one can use the svyset command to weight the sample; the command to use to weight the sample at the TA level is: svyset TA, strata(pmu) fpc(fpc1) || EAid, fpc(fpc2) || village3 || _n , singleunit(scaled) poststrata(poststratum) postweight(poststratsize)
57
Region/Stratum District Traditional Authority
Northern
Chitipa Mwaulambya
Rumphi Mwankhunikira
Mzimba Chindi
Kampingo Sibande
Mtwalo
Nkhata Bay Kabunduli
Mzuzu Viphya ward
Central
Kasungu Simlemba
Lilongwe City Area 25 ward
Area 36 ward
Dedza Pemba
Tambala
Ntcheu Kwataine
Southern
Balaka Kalembo
Blantyre Kapeni
Blantyre City Namiyango ward
Chikwawa Chapananga
Ngabu
Mangochi Jalasi
58
Mulanje Mabuka
Nsanje Mbenje
Zomba Mwambo
Table 1: Traditional authorities and local government wards included in the LGPI survey,
Malawi 2016
59
Table 2: Villages included in the LGPI survey, Malawi 2016
Northern Region
Chitipa District,
Mwabulambya TA
Mzimba District,
Mtwalo TA
Mzimba District,
Chindi TA
Beard Ngwale Nyondo Jamu Mbeye Alifeyo Mphepo
Chamanthenga Kajiso Shaba Beleji
Isaac Nyondo
Kamukwamba
Nyambose Bundi
James Nyondo
Kamzunguzgeni
Zgambo
Chimbizga
Gondwe
Kasisi 1 Katandula Chimkungule
Lazaro Chizimu Katandula Mkandawire Chimujithe
Moses Nyondo Lazaro Jere
Chitowo
Kumwenda
Mukono Siyombwe
Mahekeya Blackwell
Makwakwa Gayo
Mwakawanga Mapale Masasa Katona Jumbo
Mwalala Siyombwe Mboyonga Katuwa Nyasulu
Nankhalamu Katutula Mkumbwa Mabongo Nyirenda
Simwambi Msokwa Phiri Mdikangulu
Ten Nyondo Muthakapoli Longwe Mkandawire
Yohani Chizimu
Nyondo
Simon Chingwa
Munthali Thom Chirambo
Yotamu Nyondo 2 Sondwani Nhlema Tizamwa
Zebedia Makwakwa Vavera Bota
Zigondo Nhlema
Nkhatabay District,
Kabunduli TA
Mzimba District,
Kampingo Sibande
TA
Mzimba District,
Viphya Ward
60
Chaola Galuka Mbeya Chapola
Chimuyawi John Kaunda Gezamgomo
Chindevu Kampamayilo Juma
Chinyakula 2 Kamweko Chavula Mapale Masasa
Chipimbininga Kanyemba Shawa Masasa
Chiuta Banthu Kazuba Nkunika Mithi
Dananji Mtayamo Masiwa Soko Mziya
John Kajiso Mawelera Tembo Zongendawa
Kamkhwalala Mwanamsula Lozi
Mdachi Mwendayekha
Moseni Panganani Nyirenda
Mphande Satiel Sibande
Mweza Sitima Nkhambule
Tunduma Zawanje Nyirenda
Vimaso
Wajumpha
Yohane
Rumphi District,
Mwankhunikira TA
Chilipapa
Chimalawanthu
Chiphwantha
Gota Harawa
Julaniko
Kaidokere Munthali
Kasimba Mwatchuka
Katatawe Mzumara
Kayunga
Mkwayira Zolokere
61
Mundango
Mwanchuka
Nkhalikali A
Nkhalikali B
Nkhalikali C
Nthandala Mzumara
Vitanda
Wasambo
Central Region
Lilongwe District, Area
25 Ward
Dedza District,
Pemba TA
Ntcheu District,
Kwataine TA
Area 25A Chinthankhwa 2 Chikala
Area 25B Chipanga Gongonya
Area 25C Chitimbe Jolijo
Dzenza Gowampingo Kalazi
Kanengo Police Kabinda Kalimwayi
Lilongwe TTC Kanyimbo Kamzangaza
Kuchipala Kawere
Mawere Nachiye
Mtengowagwa Ndadzala
Sitenala Nenekeza
Tchale
Lilongwe District, Area
26 Ward
Dedza District,
Tambala TA
Kasungu District,
Simulemba TA
Biliati Chilimata Chakondwa
Chisumbi Chiumbe Chapwawa
Kafula Kachulu Chikunthu
62
Kandikole Kamgunda Chisazima
Kaondo Kanyama Dotolo
Katantha Kasisi Gideon
Mtengowagwa Kasulo Jumbo
Phwetekere Kawire Kamchocho
Kumalaya Makwenje
Kumchiza Mayira
Majiya Mgawa
Mkajenda Thomas Kamanga
Mkwenembera
Mphonde
Napwanga
Nyongo
Southern Region
Nsanje District,
Mbenje TA
Balaka District,
Kalembo TA
Chikwawa
District,
Chapananga TA
Blaiton Amini Chakumanika
Chapasuka Bonongwe Chaleka
Chimtedza Dinala Chamera
Falawo Idi Chigwata 1
Juma Kalembo Dominiko
Kankhomba Kapito Elemani
Lesitala Machemba Fulande
Maere Makuta Galonga
Minthanje Masautso Guta
Mphamba Mboga Gwada
Samuel Mpalasa Jana
63
Sinosi Mpamasi Mdyamizu
Tambo 3 Mphemba Simonzi
Tchenyela Msaliwa Timbenao
Njoka Tomasi
Saiwala
Zomba District,
Mwambo TA
Mangochi District,
Jalasi TA
Mulanje District,
Mabuka TA
Bokosi Balakasi Kapesi
Chapalaki Chande Katute
Kwindimbule Chiumba Mpasuka Magabwa
Manyungwa Kaliyapa Michenga
Misomali Kamwendo Mikundi
Namapata Kamwepe Mjojo
Nambwale Makalani Murofinyo
Nthunya Matewule Nande
Nyangu Mkuti Ngwezu
Sapali Mkweya
Somba Mlumbwa
Tambala Mosiya
Tholola
Zomba District,
Mbenje TA
Blantry District,
Namiyango Ward
Chimtedza Chilambe
Tambo 3 Maganga
Masala
Chikwawa District,
Ngabu TA
Blantry District,
Kapeni TA
Failos Kumwembe Chauwa
Jombo Ching'amba
64
Matsukambiya Chingota
Nkhwazi Mazale
Santo Mchere
Washeni Patsani
Positi Masulani
Training and Survey Team
Interviewer training took place March 15-18, 2016 in which the research team trained 56
interviewers on survey administration, experimental question administration, and tablet
computer use. Interviewers were recruited based on experience, qualifications, and languages
spoken. The survey was conducted in Chichewa, Chitumbuka, and English. Interviewers were
divided into teams of five, each with a team leader. On average, each team spent 4 days in a TA:
one day for each EA. This helped facilitate callbacks given that the teams were in a single area
for multiple days. The survey was completed over the course of 33 days; March 26, 2016 – April
27, 2016.
Post-Survey Focus Groups
After our initial analysis of the LGPI survey data and the survey experiment, we conducted FGDs
with women, men, and youth to further explore our findings and to gain a deeper understanding
of voter attitudes toward vote buying and ethnic/regional ties. Focus group discussions were
conducted with a total of 99 participants in two villages each in the two districts of Dedza and
Mzuzi. The main purpose of holding the FGDs was to receive feedback on the findings from our
conjoint experiment and LGPI findings. The discussions centred around the following main
themes: 1) Local/National Elections, 2)Vote Buying, 3) Local Origin Vs Local Residence Vs
Ethnicity, 4) Ethnicization (Importance of Ethnic Group, Obligation to help: Ethnicity Vs
Religion, Importance of working with co-ethnics, Certain groups advantageous in terms of
education, health or subsides, 5, Diverse people working together, Roles in Ensuring quality
education and health services), 6) Mixed Ethnicity.
65
A total of 12 FGD sessions—three in each of the four sample sites—were held. The highest
participation rate was ten respondents, while the minimum group had seven. Overall, a total of 99
people participated in the FGDs, 34 men and 33 women, 32 youth. Based on neighbourhood, it is
possible to divide our FGDs by income brackets. The three focus groups conducted for men,
women and youth in Masasa (Mzuzu), may be characterized as middle class where typically, people
that are able to pay bills but not save. Typically, the FGD members in Masasa were either engaged
in business or they reported to be civil servants (men). The nine other focus groups conducted in
rural Mzuzi (Malivenji village) and Dedza (Kabinda and Gunduze) can be characterised as low
income, primarily rural people dependent on subsistence farming.
Post-Survey Focus Group Discussions District Village/town
Number of focus
group participants
(youth)
Number of focus
group participants
(male)
Number of focus
group participants
(female)
Dedza
Gunduze 8 7 10
Kabinda 8 9 6
Mzuzu
Masasa 8 10 10
Malivenji 8 8 7
TOTAL 4 32 34 33